Tertiary education expansion and regional firm development

ABSTRACT

This study investigates the impact of a tertiary education expansion on regional firm development, as measured by average profits per firm. We exploit the quasi-random establishment of universities of applied sciences (UASs) – bachelor’s degree-granting three-year colleges teaching and conducting applied research – to construct treatment and control groups and to apply both a difference-in-differences model and an event study design. We find that after the establishment of new UASs in Switzerland, average profits per firm in the treated municipalities increase by 19.6% more than in the control group. This increase corresponds roughly to an additional annual growth in average profits per firm in the treatment group of 0.7%. The effects start shortly after the establishment of UASs but also persist over a period of up to 10 years.

KEYWORDS:

• higher education and research institution
• innovation
• regional firm development
• administrative tax data

JEL:

• I23
• I26
• O18
• O30

INTRODUCTION

The theoretical literature on education economics focuses on a number of outputs of tertiary education institutions, for example, knowledge creation through research and development (R&D), human capital creation, transfer of existing know-how or capital investments (Drucker & Goldstein, 2007; Florax, 1992; Goldstein & Renault, 2004; Shaw & Allison, 1999). These different outputs have the potential to impact the regional economy in a variety of ways. Indeed, empirical studies have repeatedly proved the existence of such impacts on the regional economy (for reviews, see Brekke, 2020; Caniëls & van den Bosch, 2011; Drucker & Goldstein, 2007; Peer & Penker, 2016). These impacts include – but are not limited to – consumption increases through employees and students of tertiary education institutions (Garrido-Yserte & Gallo-Rivera, 2010), more business innovation (Andrews, 2019; Cowan & Zinovyeva, 2013; Toivanen & Väänänen, 2016), new start-ups and spin-offs (Audretsch & Lehmann, 2005; Baltzopoulos & Broström, 2013; Fritsch & Aamoucke, 2017), productivity gains (Andersson et al., 2004; Barra & Zotti, 2017; Hermannsson et al., 2017), and higher regional growth (Manca, 2012).

However, while many of these regional impacts of tertiary education institutions apply to firms, it remains unclear whether they also pay off in terms of regional firm development as measured by average profits per firm in a region. On one hand, the literature shows that innovation, as measured by patents, has a long-run positive effect on firm profits (Czarnitzki & Kraft, 2010; Ernst, 2001; Geroski et al., 1993; Kaiser, 2009) and firm growth (Grabowski & Mueller, 1978). Furthermore, economic impact assessments of tertiary education institutions show local benefits both through the direct and indirect spending of the tertiary education institution and their employees (Schubert & Kroll, 2016) and through local spillover benefits from the human capital created by the tertiary education institution (Bloom et al., 2013; Siegfried et al., 2007).

On the other hand, innovations require expensive investments that could – at least in the short run – lead to a slower development of innovative firms (Beck et al., 2016; Nunes et al., 2012; Schilling, 2017). Furthermore, as innovations replace old products, negative effects of the innovation of a particular firm on the economic development of its rival firms are possible (Bloom et al., 2013; Chun et al., 2016). More generally, Moretti (2010) argues that general equilibrium effects, such as changes in wages and rents, affect the development of all firms in a region. Higher wages and rents caused by higher average levels of human capital (e.g., Rauch, 1993) could thus hamper the development of firms that do not profit from positive impacts of tertiary education institutions. Therefore, whether and, if so, when and to what extent the impacts of a tertiary education institution have on the overall positive effects on regional firm development remain open questions.

To answer these questions, we investigate a tertiary education expansion – formerly shown to yield positive impacts on the number of patents (Pfister et al., 2021) and R&D employment (Lehnert et al., 2020) – to analyse whether these impacts have the potential to positively influence regional firm development. The educational expansion we analyse took place in Switzerland in the form of the establishment of universities of applied sciences (UASs) – bachelor’s degree-granting three-year colleges teaching and conducting applied research. The establishment of UASs is particularly useful for the purpose of our analysis for two reasons. First, as the establishment of UASs took place in a quasi-random manner (Pfister, 2017; Pfister et al., 2021), the timing and location was unpredictable for local firms and therefore did not influence firms’ earlier development (more details – both qualitative and quantitative – on the process of the establishment of UASs and the impact on the regional economy are discussed in Appendix A in the supplemental data online).

Second, by exploiting the quasi-random establishment, previous studies show positive impacts of the establishment of UASs in the form of increases in innovation, as measured by patents (Pfister et al., 2021) and increases in R&D employment shares (Lehnert et al., 2020). We apply a similar methodology as these studies to identify causal effects of the establishment of UASs. However, we build on – and further develop – their findings by studying a novel outcome variable, that is, average profits per firm, and by answering the question whether such innovation effects as found in earlier studies translate into increased average profits per firm at the regional level. We, thereby contribute to both the literature on education and regional economics and to the evaluation of a particular policy by analysing whether the many forms in which a tertiary education expansion can impact firms, pay off in terms of regional firm development as measured by average profits per firm.

To conduct our analysis, we combine two data sources. Our first dataset is data on the timing and location of the establishment of UASs in the late 1990s in Switzerland. We, thereby use the dataset on the establishment of UASs from Pfister et al. (2021), augmented by Schlegel et al. (2021), to assign municipalities to treatment and control groups.

For our second dataset, we obtained access to Swiss corporate income tax data at the municipality level provided by the Swiss Federal Tax Administration (SFTA) from 1973–2016 (latest year available). The data contain information on aggregate firm profits (i.e., the total reported taxable firm profits) and the number of taxable firms at the municipality level, thereby allowing us to calculate the average profits per firm in a municipality, our measure for regional firm development. These data are particularly adequate to analyse effects of UASs on regional firm development because of two reasons. First, the tax data represent an administrative data source, and therefore are especially reliable because they comprise the whole universe rather than a subsample of all taxable firms in Switzerland.

Second, by aggregating data at the municipality level, which constitute the smallest political entities in Switzerland, we can investigate the effects of UASs on regional firm development at a very fine-grained geographical level. This fine-grained geographical level is important because effects of tertiary education institutions are found to be geographically localized (e.g., Abramovsky & Simpson, 2011; Drucker, 2016; Rodríguez-Pose & Crescenzi, 2008). The tax data allow also a much more fine-grained geographical analysis than alternative outcome measures such as productivity or value added that are only available at the supra-regional level (i.e., the cantonal level) where, on average, around 83 municipalities are aggregated. By combining the tax dataset and the dataset on the establishment of UASs, we assess the effects of the establishment of UASs on regional firm development.

In our empirical analysis, we use in a first step a two-way fixed effects difference-in-differences (DiD) approach to exploit the quasi-random variation in the timing and location of the establishment of UASs. To study dynamic treatment effects, that is, to learn how the impact of the establishment of UASs developed over time, we apply in a second step an event study design. Further analyses shed light on the question whether the detected innovation effects by Pfister et al. (2021) moderate effects on regional firm development. A number of robustness tests, furthermore, show that our results are not driven by the characteristics of the municipalities that happen to be in the treatment and control groups.

The results of our DiD regression show that average profits per firm in treated municipalities increase on average by 19.6% more than in non-treated municipalities. Analysing the dynamic treatment effects, we find that average profits per firm increase immediately after the establishment of UASs and grow further before they stabilize in the long run. Our further analyses reveal that municipalities that profit from UASs in terms of higher innovation also show the biggest increases in average profits per firm, providing empirical evidence that this might be one channel through which the establishment of UASs impacts regional firm development.

The remainder of the study is organized as follows. In the next section we survey the literature. The third section presents the data. The fourth section discusses the empirical strategy. The fifth section presents the results, further analyses and robustness checks. The sixth section summarizes and concludes.

LITERATURE

The outputs of tertiary education institutions and, thereby, their impacts on regional economies go beyond the outputs of the core functions of tertiary education institutions, that is, to create human capital and knowledge by teaching, conducting and sharing R&D (Drucker & Goldstein, 2007; Florax, 1992; Goldstein & Renault, 2004). The left-hand side of Figure 1 summarizes the different outputs frequently discussed in the theoretical literature on education and regional economics. These literature emphasize that beyond their core functions, tertiary education institutions transfer existing knowledge, for example, by cooperating with firms (Arvanitis et al., 2006; Audretsch & Feldman, 1996) and contribute to a regional knowledge infrastructure, for example, by providing laboratories (Abel & Deitz, 2012; Hussler & Ronde, 2005; Varga, 2000). In addition, tertiary education institutions create employment and make investments (Siegfried et al., 2007). All these outputs have the potential to impact the individuals and the economy of the region with a tertiary education institution in many different ways, as shown on the right-hand side of Figure 1 (Drucker & Goldstein, 2007; Florax, 1992; Schubert & Kroll, 2016).

Figure 1. Outputs and expected regional economic impacts of tertiary education institutions.

Sources: Authors based on Goldstein et al. (1995) and Goldstein and Renault (2004).

Many impacts of tertiary education institutions are extensively studied in the literature on education and regional economics. Improved human capital accelerates productivity in the local labour market (Andersson et al., 2004; Barra & Zotti, 2017; Glaeser et al., 2014; Manca, 2012), and existing knowledge spills over to local firms (Valero & Van Reenen, 2019). Other impacts, detected by the empirical literature to be linked to the proximity to tertiary education institutions, are increases in patents (Jaffe, 1989; Toivanen & Väänänen, 2016; Valero & Van Reenen, 2019) – with effect sizes between 7% (Cowan & Zinovyeva, 2013) and 32% (Andrews, 2019) in higher patenting activities – increased R&D (Abramovsky & Simpson, 2011), and more start-ups (Audretsch & Lehmann, 2005; Bathelt et al., 2011; Fritsch & Aamoucke, 2017). More generally, a tertiary education institution can help to develop a regional knowledge infrastructure that eventually leads to agglomeration economies (Abel & Deitz, 2012; Hussler & Ronde, 2005; Varga, 2000).

The occurrence of these impacts of tertiary education institutions have the potential to create further effects on the entire regional economy (Drucker & Goldstein, 2007; Florax & Folmer, 1992). The classical examples of such effects are an acceleration of gross domestic product (GDP) growth or a decrease in unemployment due to changes in the regional economy induced by a tertiary education institution (Barra & Zotti, 2017; Hermannsson et al., 2017; Schubert & Kroll, 2016). A number of empirical studies, therefore, find a positive impact on regional GDP per capita growth rates of increased human capital (measured by the regional share of individuals with tertiary education) in general (Abel & Gabe, 2011; Barra & Zotti, 2017; Chatterji, 1998; Cuaresma et al., 2014) and of tertiary education institutions in particular (Chatterji, 1998; Pastor et al., 2018).

However, less is known about how the impacts of tertiary education institutions influence the economic development of firms in a region (hereinafter, we refer to these effects as ‘regional firm development’). Therefore, in this study we ask and answer the question of whether the impacts of a tertiary education expansion have an overall effect on regional firm development as measured by the average profits per firm in a region nearby a newly established tertiary education institution in comparison to regional firm development in regions without a newly established tertiary education institution. As tertiary education institutions can affect average profits per firm both positively or negatively, it is empirically not per se clear whether the impacts of a tertiary education expansion translate into a positive or negative effect on regional firm development and how large these total effects might be. The question regarding the direction of the overall effects of tertiary education institutions on regional average profits per firm thus remains an empirical question that we tackle in this paper.

On the one hand, three empirical arguments for different positive effects on regional average profits per firm exist. First, empirical analyses on the link between patenting and firm performance show that innovations increase firm profits and positively impact firms’ performance (as measured by sales), respectively (Czarnitzki & Kraft, 2010; Ernst, 2001; Geroski et al., 1993). Analysing Danish firms, Kaiser (2009) finds that a one unit increase in the number of patents increases profit rates (i.e., profits/sales) two years later by 6.9%. Kaiser (2009) furthermore points to the facts that firms need time for the transfer of patents to profitable products. Thus, also profits from innovation effects of a tertiary education institution need time to evolve.

Second, positive innovation effects of a tertiary education institution on one firm can lead to technology spillovers to other firms both within and across industries (Hall et al., 2010; Jones, 2005; Keller, 2004). Analysing such spillovers, Bloom et al. (2013) find that the social returns to R&D (i.e., the aggregate output increase after one firm marginally increased its R&D) are twice as high as private returns to R&D (i.e., the increase in the output of the firm that increased its R&D).

Third, and more generally, Moretti (2010) argues that increases in the demand for local goods and services – an effect that is also observed in the context of tertiary education institutions (Siegfried et al., 2007) – can positively influence regional economic development through regional multiplier effects. Such multiplier effects, however, not only apply to demand-side effects but also indirectly to supply-side effects (when, for example, leading to more innovation).

On the other hand, impacts of tertiary education institutions – particularly impacts on innovation – can have negative effects on regional average firm profits, for the following three reasons. First, innovation is costly (Beck et al., 2016; Nunes et al., 2012; Schilling, 2017) – in particular when the innovation stems from a co-invention or an acquisition of a patent of a tertiary education institution (Cerulli et al., 2021). Therefore, positive innovation effects linked to tertiary education institutions can (in the short run) lead to a reduction of regional firm development. Kaiser (2009), for instance, finds a short-run effect of a 4.92% decrease of a firm’s profitability when the number of patents increased in the previous year.

Second, innovation provokes ‘product market rivalry effects’ (Bloom et al., 2013), potentially also for innovation spillovers from tertiary education institutions. Innovative products by some firms can, thereby slow down the development of ‘technology losers’ in a region (Chun et al., 2016). However, evidence for the existence of product market rivalry effects is mixed. Bloom et al. (2013), for instance find that positive effects from technology spillovers outweigh product market rivalry effects. In turn, Hanel and St-Pierre (2002) estimate negative effects of R&D activities by competitors on the profitability of a firm, even if these activities lead to quasi-public knowledge. Furthermore, while direct market rivalry effects on private firms through R&D conducted by academic universities are limited because they compete only in very few markets (Koch & Simmler, 2020), such negative effects are potentially more pronounced in the case of UASs that (1) conduct more applied R&D, (2) cooperate more often with private firms and (3) also provide more consulting activities than academic universities (Arvanitis et al., 2008).

Third, tertiary education institutions can also initiate general equilibrium effects such as changes in wages and housing prices (Rauch, 1993) that affect the development of all firms in a region (Glaeser et al., 2001). This applies especially to firms in the tradeable goods sector, where increasing regional prices lower the competitiveness on supra-regional markets.

In sum, whether the impact of a tertiary education expansion on factors like innovation, R&D and regional demand have an overall positive effect on the aggregated regional firm development in regions with newly established tertiary education institutions remains unclear. Answering this question and estimating the size of such an effect thus remains primarily an empirical question. The goal of our paper is to close this research gap by a thorough causal analysis of the effects of newly established UASs in Switzerland.

VARIABLES, DATA AND DESCRIPTIVE STATISTICS

To investigate our research question, we merge two datasets: first, data on the timing and location of the establishment of UASs and, second, data on regional firm development as measured by average profits per municipality.

Our first dataset, the data on the establishment of UASs, contains detailed information on the timing and the location of the process of the establishment of UASs from 1997 (the first establishment year) to 2011. These data – collected by Pfister et al. (2021) and Schlegel et al. (2021) – contain information on the establishment of all campuses where science, technology, engineering and mathematics (STEM) fields are taught.¹ The establishment data allows us to form treatment and control groups. We assign municipalities to the treatment group when they are within a 25-km radius from a STEM-teaching UAS campus. In so doing, we follow Pfister et al. (2021), who choose the 25-km radius because it corresponds to the maximum commuting distance of more than 90% of Swiss workers. Figure 2 depicts the UAS campuses and the resulting treatment and control groups formed from the 2222 Swiss municipalities.

Figure 2.  Universities of applied sciences (UAS) campuses in science, technology, engineering and mathematics (STEM) fields, establishment years and assignment to treatment and control groups.

Note: Shown are all 22 UAS campuses with STEM fields established between 1997 and 2003. The different establishment years are indicated by differently shaped markers. The 2222 municipalities are assigned to the treatment group (dark colour) if they are within a 25-km radius from a newly established UAS and otherwise to the control group (lighter colour). Numbers indicate the exact location of each UAS campus in STEM fields: (1) Bern (closed in 2003), (2) Biel/Bienne, (3) Burgdorf, Fribourg, (5) Horw, (6) Manno, (7) Muttenz, (8) Brugg, (9) Le Locle, (10) Oensingen (closed in 2003), (11) Saint-Imier, (12) Sion, (13) Wädenswil, (14) Winterthur, (15) Yverdon-les-Bains, (16) Zürich, (17) Chur, (18) St. Gallen, (19) Buchs (SG), (20) Rapperswil-Jona, (21) Genève and (22) Olten (closed in 2005).

Sources: Authors based on Pfister et al.’s (2021) and Schlegel et al.’s (2021) data.

Our second dataset, the data on regional firm development, contains information on aggregate firm profits subject to taxes at the municipality level and is provided by the SFTA.² We use the data to construct our dependent variable, the average profits per firm and a number of control variables at the municipality level. The data are available for tax periods 1971–2016 (the latest year available). For 1971–95, tax data are available for two-year periods, as taxes were assessed only every second year. Therefore, the data represent values averaged over two years. We simply assign the data reported to the first year of this two-year period, for example, the data for 1971/72 are assigned to 1971. After 1995, tax data are available yearly. Furthermore, data for some years (1987, 1993 and 1995–97) are missing for all municipalities in the dataset due to a lack of data quality (SFTA, 2016).

To construct our dependent variable, the average profits per firm, we use the aggregate firm profits subject to taxes at the municipality level and divide it by the number of taxable firms. The calculation of firm profits subject to federal taxes is standardized across Switzerland, making our dependent variable comparable across Swiss municipalities (for a discussion on how the SFTA calculates firm profits subject to taxes, see Appendix B1 in the supplemental data online). Descriptive statistics for our dependent variable and the data used to construct it appear in Table B4 in Appendix B1 online. Furthermore, Figure 3 shows the evolution of our dependent variable for treatment (triangle) and control (circle) groups. It gives descriptive evidence that average profits per firm start to increase stronger in the treatment group after the establishment of UASs, as compared with the control group. At the same time, average profits per firm evolve similar in treatment and control groups before the establishment of UASs.

Figure 3. Development of average profits per firm in treatment and control groups.

Note: Shown is the evolution of average profits per firm for treatment (triangle) and control (circle) groups. Where no lines are indicated, data on average profits per firm are missing for all municipalities in the corresponding year. Non-solid lines indicate 95% confidence intervals. The shaded area indicates the years during which new universities of applied sciences (UASs) were established. The graph reveals that differences in average profits per firm are not statistically significant between treatment and control groups before the establishment of UASs. However, they do become so after the establishment of UASs.

The tax data provide two more municipality-level variables, which allow us to control for potential regional differences in reported average profits per firm (i.e., regional differences linked to the complex Swiss taxation system). Depending on the legal form of the firm (e.g., corporations or limited liability companies), the profits subject to taxation can differ (for details, see Appendix B1 in the supplemental data online). We therefore control for the ‘average corporate tax rate’ in percentages and the ‘participation exemption’ in Swiss francs (CHF). Both variables contain information on the mixture of legal types of firms and thus partly control for different municipality requirements for firm profit taxation.

Furthermore, we control for tax reforms at the cantonal level. We do so, since Galletta and Redonda (2017) show that cantonal flat tax reforms on corporate profit taxes have negative impacts on businesses’ investment decisions. This impact, in turn, is likely to also influence average profits per firm. Therefore, it is crucial to control for this potential impact on average profits per firm through cantonal tax reforms. In Appendix B3 in the supplemental data online, we discuss our strategy to control for cantonal tax reforms in detail.

Combining the data on the establishment of UASs with the data on regional firm development, we study the period 1971–2008. While the starting point is determined by data availability, we limit our analysis of this tertiary education expansion to 2008 to avoid biases potentially induced by a later educational expansion, the introduction of master’s degree programmes (Pfister et al., 2021). In total, we have 21 periods in which data are available between 1971 and 2008.

EMPIRICAL STRATEGY

To examine whether the establishment of UASs has an effect on regional firm development, we apply in a first step a generalized two-way fixed effects DiD approach (i.e., time and unit fixed effects). We assume a treatment lag of three years, because all of the potential impacts through which the UASs could influence regional firm development – positively or negatively – need time to evolve. In so doing, we follow Pfister et al. (2021). They argue that the establishment of UASs affects regional human capital as one potential source of innovation impacts with a lag of three years – the typical time needed for students to finish a bachelor’s programme at a UAS.

The main specification of our DiD regression equation has thus the following form with average profits per firm in municipality i, canton k and year t as our dependent variable, that is, $y_{ikt}$: (1) $y_{ikt}=\delta Trea{t}_i\ast Pos{t}_{i\left(t-3\right)}+{\mathit{x}}_{\mathit{i}t}^{\mathbf{\prime }}\text{β}+{\alpha }_i+{\lambda }_{kt}+{ϵ}_{it},$(1) where $Trea{t}_i$ is a dummy variable that indicates whether a municipality i is in the treatment group; and $Pos{t}_{i\left(t-3\right)}$ is a dummy variable that indicates whether municipality i is actually treated in year t, that is, a nearby UAS was established at least three years ago. The main interest of the regression analysis applies to the coefficient $\delta$ that represents the lagged effect of the establishment of a UAS on the average profits per firm, that is, the treatment effect. Furthermore, ${\mathit{x}}_{\mathit{i}t}^{\mathbf{\prime }}$ is a vector that contains the two control variables (1) average federal corporate tax rate and (2) the participation exemption. We include municipality fixed effects $\left({\alpha }_i\right)$ to control for time-invariant municipality-specific characteristics. Furthermore, we also include canton-specific time trends (${\lambda }_{kt}$) and thereby follow earlier studies analysing the Swiss tax system (e.g., Galletta & Redonda, 2017). Including canton-specific time trends accounts for policy changes with respect to cantonal tax laws implemented over time that might also affect average firm profits.

In a second step, we extend the two-way fixed effects DiD by exploiting the staggered setting of the establishment of UASs, and adopt an event study design. This exercise provides us, on the one hand, with information on the pre-treatment trends that should run parallel for both treatment and control groups. This is also the key identifying assumption for our two-way fixed effects DiD analysis in equation (1). The event study thus provides us with ‘a test for causality’ (Angrist & Pischke, 2009, p. 237). Thereby, the event study is particularly appropriate to analyse pre-treatment trends because it allows for the staggered manner of the establishment of UASs and thus depicts pretreatment trends for all establishment years from 1997 to 2003.

On the other hand, the event study allows us to examine year-by-year treatment effects to shed more light on the question of how much time is needed from the establishment of a UAS to the impact on regional firm development. Following the notation used by Schmidheiny and Siegloch (2020), our event study specification is given by equation (2), again with average profits per firm in municipality i, canton k and year t as our dependent variable, that is, $y_{ikt}$: (2) $y_{ikt}=\sum _{j=\underset{\_}{j}}^{\overline{j}}{\delta }_j{b}_{it}^j+{\mathit{x}}_{\mathit{i}t}^{\mathbf{\prime }}\text{β}+\tau D_{kt}+{\alpha }_i+{\lambda }_t+{ϵ}_{it},\mathrm{w}\mathrm{i}\mathrm{t}\mathrm{h}$(2) $b_{it}^j\{\begin{array}{c}\sum _{s=\underset{\_}{t}-\overline{e}}^{\underset{\_}{j}}{d}_{i,t-s}ifj=\underset{\_}{j}\\ d_{i,t-j}if\underset{\_}{j}<j<\overline{j}\\ \sum _{s=\overline{j}}^{\overline{t}-\underset{\_}{e}}{d}_{i,t-s}ifj=\overline{j},\end{array}$where $\underset{\_}{j}$= –10 and $\overline{j}$ = 10 define the restricted event window, $\underset{\_}{t}$ = 1971 and $\overline{t}$ = 2008 the observation window, and $\underset{\_}{e}$ = 1997 and $\overline{e}$ = 2003 the years of the first and last events (i.e., the establishment of UASs), respectively.

The coefficient $b_{it}^j$ in equation (2) is thus a binned event indicator, since all observations outside the event window are binned in the maximum leads (i.e., $b_{it}^{\overline{j}}$) and lags (i.e., $b_{it}^{\underset{\_}{j}}$), respectively. Within the event window, however, $b_{it}^j=d_{i,t-j}$ is the event indicator that switches to 1 for the year of the treatment and is zero otherwise. Therefore, the coefficients of interest are the ${\delta }_j$ that represent the dynamic treatment effects for the j periods before (j < 0) or after (j ≥ 0) the event, relative to our reference period ${\delta }_{-1}$ (Schmidheiny & Siegloch, 2020). The estimation includes municipality fixed effects $\left({\alpha }_i\right)$, year fixed effects (${\lambda }_t$), and the same set of control variables $x_{it}$ as in equation (1). Furthermore, we include a dummy variable indicating all years following a cantonal flat tax reform (see Appendix B3 in the supplemental data online). We, instead, refrain from controlling for cantonal tax reforms using canton-specific time trends – as we do in our two-way fixed effects DiD – to avoid under-identification when including the additional period specific dummies in the event study.

Regarding the implementation of the event study, our dataset shows two complications as to a standard event study design. First, data for 1971–91 are based on two-year averages, while data after 1997 are based on yearly averages. For the years with two-year averages, we assign the averaged value of the outcome variable and the two control variables to both years. Thereby, we can ensure that the event indicator d_i,t−j is identified for each lag.³ Second, we face missing observations in the dependent variable for 1992–97 for all municipalities. Combined with the fact that most UASs were established in 1997 or 1998, however, this means that the event indicators for the last four pre-treatment periods, that is, j $\in$ [−4, −1], are based on relatively few observations, that is, municipalities that received the treatment between 2000 and 2003. The post-treatment periods, however, are identified based on the full sample, since after 1997, we have yearly and non-missing data.

To base our event indicators on sufficiently many observations we furthermore restrict the event window.⁴ Since treatment is staggered over 1997–2003, we have a smaller number of observations at both ends of the observation window (1971–2008). We restrict the event window to 10 pre-treatment periods since missing data in 1985 (for all municipalities) again lead to a low number of observations in those pre-treatment periods. Anticipation effects prior to 1985 are also unlikely to affect our estimation. The restriction to 10 post-treatment periods is driven by the fact that for the last year of the observation window, that is, 2008, we would also have only relatively few observations, that is, only those municipalities that were treated in 1997 (first establishment of UASs). We thus summarize pre- and post-treatment periods outside the event window in bins at the beginning (i.e., $b_{it}^{\underset{\_}{j}}$) and end ($b_{it}^{\overline{j}}$) of the event window.

We estimate equations (1) and (2) using a Poisson pseudo-maximum likelihood (PPML) estimator.⁵ Thereby, we tackle two problems raised by Santos Silva and Tenreyro (2006) that standard ordinary least squares (OLS) regression would entail. First, using a log-linearization of the empirical model in the presence of heteroskedasticity leads to inconsistent OLS estimates of the semi-elasticity. Second, the log-linearized OLS model is not defined for observations with zero values in the dependent variable. As our dataset has both of these characteristics, we use a PPML estimator as suggested by Santos Silva and Tenreyro (2006).⁶

RESULTS

Difference-in-differences (DiD) and event study

To assess the impact of the establishment of UASs on average profits per firm, we report, in a first step, the results from our two-way fixed effects DiD analysis in Table 1, estimated using PPML. Moving from left to right in Table 1, we append the base model without any controls in column (1) step-wise, with control variables, that is, the average corporate tax rate and the participation exemption in column (2), with canton-specific time trends in column (3), with labour market region fixed effects in column (4), and, finally, with municipality fixed effects in our preferred DiD specification in column (5). The coefficient for $Trea{t}_i$ shows the level difference between treatment and control groups, which is insignificant throughout all model specifications (and omitted with municipality fixed effects). The coefficient $Trea{t}_i\ast Pos{t}_{i\left(t-3\right)}$ shows our coefficient of interest, that is, the treatment effect of a newly established UAS, which is significant throughout all model specifications.

Table 1. Regression results with average profits per firm as the dependent variable.

Dependent variable: Average profits per firm	(1)	(2)	(3)	(4)	(5)
$Trea{t}_i\ast Pos{t}_{i\left(t-3\right)}$	0.5420***	0.4059***	0.2337***	0.2366***	0.1790***
	(0.0391)	(0.0498)	(0.0727)	(0.0688)	(0.0622)
$Trea{t}_i$	0.0765	0.1085	0.0164	−0.0636
	(0.0734)	(0.0737)	(0.0807)	(0.1114)
Average corporate tax rate	No	Yes	Yes	Yes	Yes
Participation exemption	No	Yes	Yes	Yes	Yes
Canton-specific time trends	No	No	Yes	Yes	Yes
Regional labour market FE	No	No	No	Yes	No
Municipality FE	No	No	No	No	Yes
Observations	45,066	45,066	45,066	45,066	45,066
(Pseudo-) R^2	0.033	0.058	0.157	0.235	0.758

Note: Results report coefficients from Poisson pseudo-maximum likelihood (PPML) regressions. Dependent variable: Average profits per firm. Robust standard errors are clustered at the municipality level. When municipality fixed effects are included (column 5), 68 municipalities are dropped from the estimation because they are separated by a fixed effect (Correia et al., 2019).

Levels of significance: *p < 0.10, **p < 0.05, ***p < 0.01.

FE, fixed effects.

Including control variables and year fixed effects lowers the treatment effect from 0.5420 in column (1) to 0.2337 in column (3) by accounting for, first, the higher average profits per firm in municipalities with higher corporate tax rates and a larger participation exemption (see Appendix B1 in the supplemental data online); and second, the strong time trend in average profits per firm (Figure 3). While the inclusion of regional labour market fixed effects in column (4) leaves the estimated treatment effect unchanged (as compared with column 3), when including municipality fixed effects in column (5), the treatment effect decreases to 0.1790, controlling away time-invariant unobservable municipality characteristics that impact average profits per firm.

Thus, looking at our preferred model in column (5), that contains the most extensive set of control variables, we find that the increase in average profits per firm is, on average 19.6% (0.1790) higher for municipalities in the treatment group after the establishment of a nearby UAS as compared with a municipality in the control group. This positive effect of the establishment of UASs on average profits per firm thus shows that, overall, the impacts on the regional economy due to the establishment of UASs also translate into significant and substantial effects on average firm profits.

To investigate the robustness of our DiD results with respect to municipality characteristics that could influence the location of newly established UASs, we conduct a number of robustness tests in Appendix D in the supplemental data online. We (1) investigate whether our results are driven by the composition of our treatment and control groups, (2) use only the variation in the timing of the treatment to identify treatment effects, (3) analyse the effect of municipalities with different regional characteristics and (4) run an inverse probability of treatment weighting regression. We find that none of these robustness tests structurally change our main DiD results.

In a second step, we allow for dynamic treatment effects by applying an event study design. The results of the event study appear in Figure 4a, with negative values on the x-axis indicating pre-treatment periods and positive values on the x-axis indicating post-treatment periods. The y-axis shows the estimated treatment effects for average profits per firm in each event period. Figure 4a reveals four findings. First, the coefficients in the event study are less precisely estimated than in the DiD model. Therefore, we are only able to detect few coefficients that are significant at conventional significance levels (e.g., the coefficients for t = 0 and 3). This finding is in line with a recent and ongoing debate on the efficiency of such methods to estimate causal effects of a treatment that has a staggered rollout (e.g., Callaway & Sant’Anna, 2020; de Chaisemartin & D’Haultfoeuille, 2020; Roth & Sant’Anna, 2021; Sun & Abraham, 2020). In sum, there is a consent that traditional event study designs suffer from too wide confidence intervals.

Figure 4. Evolution of treatment effects.

Note: (a) Dynamic treatment effects for average profits per firm. Negative values on the x-axis show the years before the treatment, t = 0 corresponds to the year a university of applied sciences (UAS) was established and the positive values on the x-axis belong to the post-treatment period. Depicted results are coefficients from Poisson pseudo-maximum likelihood (PPML) regressions, with t = −1 being the base year. Vertical lines mark confidence intervals. (b) Number of observations for each pre- and post-treatment period for estimating dynamic treatment effects. The few observations immediately before the treatment explain the negative but insignificant pretreatment effects in (a). *We do not report the much higher number of observations in the bin for t ≤ 10 (as compared with the other periods) for clarity.

Second, the estimated coefficients for the pretreatment period are mostly close to zero (and insignificant), showing that no relevant anticipatory effects of the establishment of UASs exist in the treatment group (e.g., caused by infrastructure investments for the establishment of UASs). The unlikely existence of anticipatory effects is also supported by joint F-tests not rejecting the null hypotheses that the pretreatment periods are jointly (or even pairwise) different from zero. However, the coefficient for t = −5 shows a marked drop, as compared with all other pretreatment periods. The reason for this drop lies in the data structure for our event study design, with missing data between 1991 and 1997 for all municipalities. These data gaps lead to the fact, that the coefficient in t = −5 is estimated solely based on 1992 and 1998, years with especially low average profits per firm (as compared with the base year of the event study, that is, t = −1) in the corresponding establishment groups (UASs established in 1997 and 2003). We further discuss this issue in Appendix C2 in the supplemental data online. More generally, Figure 4b shows the observations underlying the estimation of each event period.⁷

Third, Figure 4a shows that in the post-treatment period, treatment effects start to increase immediately after the treatment. This relatively immediate response of average profits per firm to the establishment of UASs points to the existence of positive, though small, demand-side effects of UASs. Thus, UASs impact average profits per firm not only through innovation and R&D. The reasoning underlying this interpretation is that tertiary education institutions influence innovation with a lag of between two to five years (e.g., Cowan & Zinovyeva, 2013; Fischer & Varga, 2006; Pfister et al., 2021), and that innovation, as measured by patents, has a positive impact on firm profits, again with a lag of two to three years (Ernst, 2001; Kaiser, 2009).

Fourth, Figure 4a shows that after the immediate increase, positive effects on average profits per firm grow further up to period t = 3 when they nearly stabilize (again the drop in t = 7 is linked to our data structure, see Appendix C2 in the supplemental data online). Thus, while showing some fluctuations, the increases over the entire post-treatment period of up to 10 years in our dataset, give some evidence for an impact of UASs that goes beyond a mere short-term effect. This result of a persistent long-term effect is also supported by means of a joint F-test that rejects the null hypothesis (at the 10% level) that the estimated coefficients for all periods of more than three years after the establishment of UASs are equal to zero.

Further analyses

In this section we shed more light on the potential mechanisms behind the increase of average profits per firm after the establishment of UASs.⁸ We suppose that one mechanism for higher average profits is through the positive impact of the establishment of UASs on innovation activity, as shown by Pfister et al. (2021). Drawing on patent data from the PATSTAT Worldwide Patent Statistical Database – April 2013 Version, we analyse whether municipalities that profit strongly in terms of innovation (as measured by patenting activities), after the establishment of UASs also profit more strongly in terms of profits from the establishment of UASs. If so, this provides some evidence that the higher average profits after the establishment of UASs is the result of higher innovation activities (linked to the establishment of UASs) that pay off.

We thus estimate the effect of the establishment of UASs on innovation activities, using a three-year lag, and use the estimated treatment effect to group municipalities into quintiles, according to the size of the innovation effect associated with the establishment of UASs. The resulting categorical variable is then interacted with a treatment dummy, now lagged by six years, to account for the time needed for innovation to affect profits (Ernst, 2001). Figure 5 shows the result of this exercise. The higher and significant coefficients for average firm profits in the two uppermost quintiles with respect to innovation effects of UASs reveal that only municipalities that show positive impacts of UASs in the form of higher innovation, also experience positive effects in the form of higher average profits per firm some three years later. Furthermore, in Appendix E1 in the supplemental data online, we provide further evidence for the importance of innovation effects linked to UASs as potential channels through which the establishment of UASs affects average profits per firm, by showing that municipalities with a higher share of firms in innovative industries tend to experience a stronger increase in average profits per firm after the establishment of UASs than municipalities with a lower share of firms in these industries.

Figure 5. Effects on average profits per firm by size of innovation effects (as measured by patenting activities) linked to newly established universities of applied sciences (UASs).

Note: Reported are coefficients from a Poisson pseudo-maximum likelihood (PPML) regression with control variables, as shown in equation (1). Dependent variable: Average profits per firm. The vertical lines indicate confidence intervals. The regression includes an interaction of the treatment dummy with a categorical variable indicating to which quintile a municipality belongs, with respect to the innovation effect (as measured by patenting activities) after the establishment of UASs. To account for a lag between innovation effects and effects on average profits, we assume a six-year instead of a three-year lag. The regression includes fixed effects at the regional labour market level (as with municipality fixed effects, the quintile dummy would be dropped) and canton-specific time trends.

CONCLUSIONS

This study analyses the effect of the establishment of tertiary education institutions on regional firm development, using the establishment of UASs – bachelor’s degree-granting three-year colleges teaching and conducting applied research – in Switzerland as a case study. Our results show a positive and significant increase in regional firm development, as measured by average profits per firm. In treated municipalities, that is, those near a newly established UAS, these profits increase significantly more than in municipalities in the control group. The magnitude of this effect corresponds to 19.6%, leading roughly to an additional annual growth in average profits per firm in the treatment group of 0.7%. Further analyses provide evidence that the increase in average profits per firm after the establishment of UASs is moderated by innovation impacts of UASs on the regional economy. Our robustness tests show that the effects on average profits are not driven by municipality characteristics that are particular to municipalities in the treatment group.

Furthermore, the analysis of the dynamic treatment effect reveals, first, that average profits per firm react immediately to the establishment of the UASs. Second, the dynamic treatment effects show patterns of increasing average profits up to three years after the establishment and then remain stable in the long run. However, effects are often not statistically significant. The direct increase combined with the long-term persistence reveal that both, short run – rather expenditure based – and long run effects – for example, through increased innovation – are at play.

However, future research should address some of the limitations this study faces – predominantly imposed by a lack of data. First, the analysis should be expanded to UASs in non-STEM fields, since it is likely that regional firm development, as measured not only by average profits per firm, are also affected by UASs in these fields.⁹ Finding out the importance of the impact on regional firm development of UASs in all fields in general, and of each specific field in particular, would be an important future research question. Second, our setting does not allow us to exactly pin down the mechanisms through which UASs influence regional firm development. The establishment of UASs might lead to both supply and demand side effects, for example, cost-reducing product and process innovations (supply side), new but more expensive products (both, supply side, through more qualified workforce and demand side through more sophisticated consumers) or general equilibrium effects such as changes in wages and rents (demand side). Therefore, providing evidence for positive effects of UASs on regional firm development is only the first step and finding out the underlying mechanisms remains a pathway for future research.

Despite these limitations, our findings are affirmative regarding the achievement of some of the goals of the reform leading to the establishment of UASs (see Appendix A in the supplemental data online). The goals were not only to create both human capital and knowledge (the core functions of tertiary education institutions), but to foster technology transfers, increase cooperations with local firms and ‘revitalize the Swiss economy’ (Bundesrat, 1994). The positive effects on regional firm development imply that these additional goals are achieved at the aggregated municipality level.

More generally, these positive effects on regional firm development show that, overall, the impact of newly established tertiary education institutions on the regional economy eventually translate into improved regional firm development, as measured by average profits per firm. The repeatedly documented positive impact of tertiary education institutions, for example, increases in innovation and knowledge spillovers, reflect thus not only quantitative but qualitative increases, meaning that innovations and knowledge spillovers pay off on average. Furthermore, our analysis shows that these increases are not at the expense of regions without newly established tertiary education institution but reflect an additional stimulus for regional firm development.

ACKNOWLEDGEMENTS

This paper benefited from many audiences and exchanges. In particular, we thank the following for their help and support: Simone Balestra, Simon Janssen, Edward Lazear, Jens Mohrenweiser, Natalie Reid and Alexandre Ziegler. Moreover, we thank the seminar participants at the University of Zurich and conference participants at EEA 2020, SASE 2020, VfS 2020 and FHNW 2021. Furthermore, we thank Dietmar Harhoff and the SFTA for the provision of data used in our analyses.

DISCLOSURE STATEMENT

No potential conflict of interest was reported by the authors.

Additional information

Funding

This study is partly funded by the Staatssekretariat für Bildung, Forschung und Innovation (SERI – Swiss State Secretariat for Education, Research, and Innovation) through its Leading House on the Economics of Education, Firm Behaviour and Training Policies.

Notes

1. The reasons to analyse STEM campuses are that we focus particularly on innovation impacts and their influence on average profits per firm. We discuss these reasons in more detail in Appendix A in the supplemental data online.

2. The data were collected within the research project ‘The Swiss Confederation: A Natural Laboratory for Research on Fiscal and Political Decentralization’ (2010–13), directed by Marius Brülhart and which has already been used in studies in the field of economic geography (e.g., Brülhart et al., 2012; Luthi & Schmidheiny, 2014).

3. The leads are identified either way, since after 1997 we have yearly data and the establishment of UASs only starts in 1997.

4. In our particular case, binning is thus not done for identification purposes (Schmidheiny & Siegloch, 2020) because we still have the control group included in our estimation.

5. To attenuate the influence of outliers, we follow Cox (2013) and estimate the PPML estimator excluding the 0.5% biggest municipalities. We discuss this issue in detail in Appendix C1 in the supplemental data online.

6. (1) The Breusch–Pagan test rejects the null hypothesis of homoskedasticity at all conventional significance levels. (2) As Table B2 in the supplemental data online shows, our dataset has a considerable number of observations with zero spells, especially in earlier years.

7. As 16 out of the 22 UAS campuses were established between 1997 and 1998, and as data for 1991–97 are missing for all municipalities, the estimation for these pretreatment periods rests on the few observations for municipalities with UASs established after 1998.

8. In Appendix E in the supplemental data online, we discuss four further analyses. First, we analyse the impact of different pretreatment shares of innovative firms in a municipality. Second, we investigate whether effects are rather driven by increases in profits or by a reallocation of above-average profitable firms to the treatment groups. Third, we run separate analysis for the establishment of UASs as greenfield (ex nihilo) and brownfield institutions. Fourth, we investigate how the treatment effect changes with growing distance to newly established UASs and with other levels of geographical analysis.

9. Apart from STEM, that is, UASs specializing in ‘engineering and IT’ and ‘chemistry and the life sciences’, UASs also specialize in (1) ‘agriculture and forestry’; (2) ‘architecture, construction and planning’; (3) ‘music, theater and other arts’; (4) ‘design’; (5) ‘health’; (6) ‘applied linguistics’; (7) ‘business, management and services’; (8) ‘applied psychology’; (9) ‘social work’; and (10) ‘sports’.